WO2008144960A1 - Method and apparatus for mpi program optimization - Google Patents
Method and apparatus for mpi program optimization Download PDFInfo
- Publication number
- WO2008144960A1 WO2008144960A1 PCT/CN2007/001747 CN2007001747W WO2008144960A1 WO 2008144960 A1 WO2008144960 A1 WO 2008144960A1 CN 2007001747 W CN2007001747 W CN 2007001747W WO 2008144960 A1 WO2008144960 A1 WO 2008144960A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- shared
- shared data
- processes
- memory
- data
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F8/00—Arrangements for software engineering
- G06F8/40—Transformation of program code
- G06F8/41—Compilation
- G06F8/44—Encoding
- G06F8/443—Optimisation
Definitions
- MPI Message passing interface
- MPI may be a widely used programming model for high performance computing. Originally, MPI may be designed for 5 communication among nodes running a parallel program on a distributed memory system. With the development of multiprocessors (MP) running individual processes, the use of MPI may be extended to communication among processes on a shared memory system.
- MP multiprocessors
- FIG. 1 illustrates an embodiment of a MP computing platform using MPI programs.
- FIG. 2 illustrates an embodiment of a MPI program optimizer.
- Fig. 3 illustrates an embodiment of a method of MPI program optimization during the compiling stage.
- FIG. 4 illustrates an embodiment of a method of MPI program optimization during the runtime stage.
- Fig. 5 illustrated an embodiment of a method of executing optimized MPI program.
- Embodiments of the invention may be implemented in hardware, firmware, software, or any combination thereof. Embodiments of the invention may also be implemented as instructions stored on a machine-readable medium, that may be read and executed by one or more processors. A machine-readable medium may
- a machine-readable medium may include read only memory (ROM); random access memory (RAM); magnetic disk storage media; optical storage media; flash memory devices; electrical, optical, acoustical or other forms of propagated signals (e.g., carrier
- Fig. 1 shows an embodiment of a MP computing platform.
- Platform 10 may comprise a plurality of processors 11 i-11 n , a shared cache 12 and a shared memory 13, and possibly other components for data transceiving and/or data processing.
- Processors 11 i-11 n may be used to run individual processes.
- Shared cache 12 and shared memory 13 may be associated with and sharable by processors 11i-11 n .
- Shared cache 12 may be a bridge between processors 1 I 1 - 11 n and shared memory 13 and may allow instructions and data transferred from shared memory 13 to be read or written by processors 11 r 11 n at a faster speed.
- computing platform 10 may apply cache hierarchy architecture, namely, each of processors 1 11-11 n may be associated with a first level cache (not shown), a group of processors may be associated with a second level cache (not shown) and shared cache 12 may be a shared last level cache (LLC) that may be associated with all of processors 11i-11 n .
- MPI may be a de facto standard for communication among the processes running a parallel program on a shared memory system.
- the parallel program may adopt single program multiple data (SPMD) pattern, namely, all processes running on processors may use the same program but each may 5 has its own data.
- SPMD single program multiple data
- MPI standard may provide various patterns for data communication among the processes, for example, point-to-point communications or collective communications. However, no matter which communication pattern is 10 used, a complete copy of data may be transferred between processes, which may create redundant data for each process.
- FIG. 2 illustrates an embodiment of a MPI program optimizer 20.
- MPI program optimizer 20 may optimize performance of data communication between processes.
- MPI program optimizer 20 may comprise analyzer 21 , retriever 22, 15 annotator 23, library 24, and parser 25.
- Analyzer 21 may analyze a MPI program to detect if there is data from the
- MPI program sharable by a number or all of the processes running on processors 11i-11 n .
- Various applications may determine various types of data sharable by the processes. For an information extraction application that may be used to locate
- dictionary data may be sharable by the processes.
- matrix variables may be sharable by the processes.
- integer sort application that may be a part of parallel benchmark NPB, variables used for functions such as MPI_Allreduce,
- Analyzer 21 may further analyze the MPI communication pattern for the shared data.
- MPI standard may define various communication 5 patterns, such as point-to-point communications or collective communications.
- Retriever 22 may retrieve the shared data from the MPI program to provide a shared data list.
- Annotator 23 may annotate the shared data list with a shared directive. For example, annotator 23 may annotate the shared data with a directive "#pragma shmpi shared”.
- Functionalities of analyzer 21 , retriever 22 and annotator 23 may be implemented by a programmer, a complier or other devices for compiling the MPI program.
- Library 24 may identify the shared data from the MPI program based upon the shared directive. Library 24 may further allocate the shared data in shared 15 memory 13, e.g., by using inter-process communication (IPC)/shared memory
- library 24 may first assign a buffer (buffer 130 of Fig. 1 ) from shared memory 13 to locate the shared data, in which buffer 130 may comprise a plurality of storage units with consecutive logical addresses.
- Parser 25 may transform the MPI communication pattern for the shared data into single-data-copy-access pattern. More specifically, processes may read or write the single copy of the shared data from or to the dedicated buffer 130 of shared memory 13. If a process needs the shared data updated by another process, the process can read the shared data from shared memory 13 after the another process completes updating and writing the shared data into shared memory 13. If two or more processes read the same shared data, they can do it 5 simultaneously.
- Fig. 3 shows an embodiment of a method of MPI program optimization during the compiling stage.
- analyzer 21 may analyze the MPI program about its communication pattern and data sharing information. Analyzer 21 may analyze what kind of data sharable by the processes based upon the 10 program application.
- retriever 22 may retrieve the shared data from the MPI program.
- annotator 23 may annotate the shared data with a shared directive, e.g., "#pragma shmpi shared" directive.
- Fig. 4 shows an embodiment of a method of MPI program optimization during the runtime stage.
- library 24 may identify the shared data from the MPI program based upon the shared directive.
- library 24 may allocate the shared data in buffer 130 of shared memory 13.
- library 24 may map buffer 130 to the processes that may share the data. For
- library 24 may map the address of buffer 130 to the processes.
- parser 25 may transform the MPI communication patter for the shared data into the above-sated single-data-copy-access pattern.
- Fig. 5 shows an embodiment of a method of executing optimized MPI program.
- a process may run the MPI program.
- it may be determined whether the MPI program instructs to read or write the shared data. If so, the process may obtain the address of the shared data in shared memory 13 5 in block 503. The process may obtain the address of the shared data based upon the address of buffer 130 storing the shared data that may be mapped to the process by using the library routines.
- the process may read or write the shared data from or to shared memory 13.
Landscapes
- Engineering & Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Memory System Of A Hierarchy Structure (AREA)
- Devices For Executing Special Programs (AREA)
Abstract
Machine readable media, methods, apparatus and system for MPI program optimization. In some embodiments, shared data may be retrieved from a message passing interface (MPI) program, wherein the shared data is sharable by a plurality of processes. Then, the shared data may be allocated to a shared memory, wherein the shared memory is accessible by the plurality of processes. A single copy of the shared data may be maintained in the shared memory so that more than one processes of the plurality of processes can read or write the single copy of the shared data from or to the shared memory.
Description
METHOD AND APPARATUS FOR MPI PROGRAM OPTIMIZATION
BACKGROUND
[0001 ] Message passing interface (MPI) may be a widely used programming model for high performance computing. Originally, MPI may be designed for 5 communication among nodes running a parallel program on a distributed memory system. With the development of multiprocessors (MP) running individual processes, the use of MPI may be extended to communication among processes on a shared memory system.
BRIEF DESCRIPTION OF THE DRAWINGS
[0O02] The invention described herein is illustrated by way of example and not by way of limitation in the accompanying figures. For simplicity and clarity of illustration, elements illustrated in the figures are not necessarily drawn to scale. For example, the dimensions of some elements may be exaggerated relative to other elements for clarity. Further, where considered appropriate, reference labels 15 have been repeated among the figures to indicate corresponding or analogous elements.
[0003] Fig. 1 illustrates an embodiment of a MP computing platform using MPI programs.
[0004] Fig. 2 illustrates an embodiment of a MPI program optimizer.
[Q005] Fig. 3 illustrates an embodiment of a method of MPI program optimization during the compiling stage.
[0006] Fig. 4 illustrates an embodiment of a method of MPI program optimization during the runtime stage.
[0007] Fig. 5 illustrated an embodiment of a method of executing optimized MPI program.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0008] The following description describes techniques for MPI program
5 optimization. In the following description, numerous specific details such as logic implementations, pseudo-code, means to specify operands, resource partitioning/sharing/duplication implementations, types and interrelationships of system components, and logic partitioning/integration choices are set forth in order to provide a more thorough understanding of the current invention. 10 However, the invention may be practiced without such specific details. In other instances, control structures, gate level circuits and full software instruction sequences have not been shown in detail in order not to obscure the invention. Those of ordinary skill in the art, with the included descriptions, will be able to implement appropriate functionality without undue experimentation.
[QΘ09] References in the specification to "one embodiment", "an embodiment", "an example embodiment", etc., indicate that the embodiment described may include a particular feature, structure, or characteristic, but every embodiment may not necessarily include the particular feature, structure, or characteristic. Moreover, such phrases are not necessarily referring to the same embodiment. Further,
20 when a particular feature, structure, or characteristic is described in connection with an embodiment, it is submitted that it is within the knowledge of one skilled in the art to effect such feature, structure, or characteristic in connection with other embodiments whether or not explicitly described.
[0010] Embodiments of the invention may be implemented in hardware, firmware, software, or any combination thereof. Embodiments of the invention may also be implemented as instructions stored on a machine-readable medium, that may be read and executed by one or more processors. A machine-readable medium may
5 include any mechanism for storing or transmitting information in a form readable by a machine (e.g., a computing device). For example, a machine-readable medium may include read only memory (ROM); random access memory (RAM); magnetic disk storage media; optical storage media; flash memory devices; electrical, optical, acoustical or other forms of propagated signals (e.g., carrier
10 waves, infrared signals, digital signals, etc.) and others.
[0011] Fig. 1 shows an embodiment of a MP computing platform. Platform 10 may comprise a plurality of processors 11 i-11n, a shared cache 12 and a shared memory 13, and possibly other components for data transceiving and/or data processing.
[0Θ12] Processors 11 i-11 n may be used to run individual processes. Shared cache 12 and shared memory 13 may be associated with and sharable by processors 11i-11n. Shared cache 12 may be a bridge between processors 1 I 1- 11n and shared memory 13 and may allow instructions and data transferred from shared memory 13 to be read or written by processors 11 r11 n at a faster speed.
[2013] It should be appreciated that computing platform 10 may apply cache hierarchy architecture, namely, each of processors 1 11-11 n may be associated with a first level cache (not shown), a group of processors may be associated with a second level cache (not shown) and shared cache 12 may be a shared last level cache (LLC) that may be associated with all of processors 11i-11n.
[0014] MPI may be a de facto standard for communication among the processes running a parallel program on a shared memory system. Conventionally, the parallel program may adopt single program multiple data (SPMD) pattern, namely, all processes running on processors may use the same program but each may 5 has its own data. Therefore, multiple copies of the data may be maintained in the memory system, which could be a big challenge for memory bandwidth and cache performance. Conventionally, MPI standard may provide various patterns for data communication among the processes, for example, point-to-point communications or collective communications. However, no matter which communication pattern is 10 used, a complete copy of data may be transferred between processes, which may create redundant data for each process.
[0015] Fig. 2 illustrates an embodiment of a MPI program optimizer 20. MPI program optimizer 20 may optimize performance of data communication between processes. MPI program optimizer 20 may comprise analyzer 21 , retriever 22, 15 annotator 23, library 24, and parser 25.
[0016] Analyzer 21 may analyze a MPI program to detect if there is data from the
MPI program sharable by a number or all of the processes running on processors 11i-11n. Various applications may determine various types of data sharable by the processes. For an information extraction application that may be used to locate
20 boundaries of sentences and extract name entities, dictionary data may be sharable by the processes. For an MFoId algorithm application that may be used to predict Ribonucleic acid (RNA) secondary structure, matrix variables may be sharable by the processes. For an integer sort application that may be a part of
parallel benchmark NPB, variables used for functions such as MPI_Allreduce,
MPI_Alltall and MPI_Alltoallv may be sharable by the processes. [0017] Analyzer 21 may further analyze the MPI communication pattern for the shared data. As stated above, MPI standard may define various communication 5 patterns, such as point-to-point communications or collective communications. [0018] Retriever 22 may retrieve the shared data from the MPI program to provide a shared data list. Annotator 23 may annotate the shared data list with a shared directive. For example, annotator 23 may annotate the shared data with a directive "#pragma shmpi shared". [QOI 9] Functionalities of analyzer 21 , retriever 22 and annotator 23 may be implemented by a programmer, a complier or other devices for compiling the MPI program. [0020] Library 24 may identify the shared data from the MPI program based upon the shared directive. Library 24 may further allocate the shared data in shared 15 memory 13, e.g., by using inter-process communication (IPC)/shared memory
(shm) routines. For data allocation, library 24 may first assign a buffer (buffer 130 of Fig. 1 ) from shared memory 13 to locate the shared data, in which buffer 130 may comprise a plurality of storage units with consecutive logical addresses.
Then, library 24 may map a memory address of buffer 130 to the processes that 20 may use the shared data. Compared with conventional data replication, only one copy of the shared data may be maintained in shared memory 13. [0021] Parser 25 may transform the MPI communication pattern for the shared data into single-data-copy-access pattern. More specifically, processes may read or write the single copy of the shared data from or to the dedicated buffer 130 of
shared memory 13. If a process needs the shared data updated by another process, the process can read the shared data from shared memory 13 after the another process completes updating and writing the shared data into shared memory 13. If two or more processes read the same shared data, they can do it 5 simultaneously.
[0022] Fig. 3 shows an embodiment of a method of MPI program optimization during the compiling stage. In block 301 , analyzer 21 may analyze the MPI program about its communication pattern and data sharing information. Analyzer 21 may analyze what kind of data sharable by the processes based upon the 10 program application. In block 302, retriever 22 may retrieve the shared data from the MPI program. In block 303, annotator 23 may annotate the shared data with a shared directive, e.g., "#pragma shmpi shared" directive.
[0023] It should be appreciated that the above method can be implemented in other stages.
[QΘ24] Fig. 4 shows an embodiment of a method of MPI program optimization during the runtime stage. In block 401 , library 24 may identify the shared data from the MPI program based upon the shared directive. In block 402, library 24 may allocate the shared data in buffer 130 of shared memory 13. In block 403, library 24 may map buffer 130 to the processes that may share the data. For
20 example, library 24 may map the address of buffer 130 to the processes. In block 404, parser 25 may transform the MPI communication patter for the shared data into the above-sated single-data-copy-access pattern.
[0025] It should be appreciated that the above method can be implemented in other stages.
[0026] Fig. 5 shows an embodiment of a method of executing optimized MPI program. In block 501 , a process may run the MPI program. In block 502, it may be determined whether the MPI program instructs to read or write the shared data. If so, the process may obtain the address of the shared data in shared memory 13 5 in block 503. The process may obtain the address of the shared data based upon the address of buffer 130 storing the shared data that may be mapped to the process by using the library routines. In block 504, the process may read or write the shared data from or to shared memory 13.
[0027] In order to avoid data conflict, different processes may read the shared
10 data from shared memory 13 simultaneously. However, if a process needs the shared data that is updated by another process, the process can read the shared data after the another process completes updating and writing the shared data into shared memory 13. Various mechanisms may be used to ensure that the data is accessed appropriately, for example, but not limited to, exclusive mutual access
15 mechanisms.
[0028] Although the current invention has been described in conjunction with certain embodiments, it shall be understood that modifications and variations may be resorted to without departing from the spirit and scope of the invention as those skilled in the art readily understand. Such modifications and variations are
20 considered to be within the scope of the invention and the appended claims.
Claims
1. A machine-readable medium comprising a plurality of instructions that in response to being executed result in a system: retrieving shared data from a message passing interface (MPI) program, wherein the shared data is sharable by a plurality of processes; allocating the shared data to a shared memory, wherein the shared memory is accessible by the plurality of processes; and maintaining a single copy of the shared data in the shared memory so that more than one processes of the plurality of processes can read or write the single copy of the shared data from or to the shared memory.
2. The machine-readable medium of claim 1 , wherein the plurality of instructions further result in the system: annotating the shared data with a shared directive.
3. The machine-readable medium of claim 1 , wherein the plurality of instructions further result in the system: indicating a MPI communication pattern for the shared data that the more than one process of the plurality of processes can read or write the single copy of the shared data from or to the shared memory.
4. The machine-readable medium of claim 1 , wherein the plurality of instructions that result in the system allocating, further result in the system: allocating the shared data in a buffer of the shared memory; and mapping a memory address of the buffer to the plurality of processes.
5. The machine-readable medium of claim 1 , wherein the plurality of instructions further result in the system: determining to read or write the shared data from or to the shared memory; obtaining a memory address of the single copy of the shared data in the shared memory; and reading or writing the single copy of the shared data from or to the memory address.
6. A system, comprising: a retriever to retrieve shared data from a message passing interface (MPI) program, wherein the shared data is sharable by a plurality of processes; and a library to allocate the shared data to a shared memory and to maintain a single copy of the shared data in the shared memory so that more than one processes of the plurality of processes can read or write the single copy of the shared data from or to the shared memory, wherein the shared memory is accessible by the plurality of processes.
7. The system of claim 6, further comprising an annotator to annotate the shared data with a shared directive.
8. The system of claim 6, further comprising a parser to indicate the MPI communication pattern for the shared data that the more than one process of the plurality of processes can read or write the single copy of the shared data from or to the shared memory.
9. The system of claim 6, wherein the library further allocates the shared data in a buffer of the shared memory; and maps a memory address of the buffer to the plurality of processes.
10. The system of claim 6, further comprising the plurality of processes, wherein a process of the plurality of processes determines to read or write the shared data from the shared memory; obtains a memory address of the single copy of the shared data in the shared memory; and reads or writes the single copy of the shared data from or to the memory address.
11. A method: retrieving shared data from a message passing interface (MPI) program, wherein the shared data is sharable by a plurality of processes; allocating the shared data to a shared memory, wherein the shared memory is accessible by the plurality of processes; and maintaining a single copy of the shared data in the shared memory so that more than one processes of the plurality of processes can read or write the single copy of the shared data from or to the shared memory.
12. The method of claim 11 , further comprising annotating the shared data with a shared directive.
13. The method of claim 11 , further comprising indicating a MPI communication pattern for the shared data that the more than one process of the plurality of processes can read or write the single copy of the shared data from or to the shared memory.
14. The method of claim 11 , wherein the allocating further comprises: allocating the shared data in a buffer of the shared memory; and mapping a memory address of the buffer to the plurality of processes.
15. The method of claim 11 , further comprising: determining to read or write the shared data from or to the shared memory; obtaining a memory address of the single copy of the shared data in the shared memory; and reading or writing the single copy of the shared data from or to the memory address.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US12/598,597 US8312227B2 (en) | 2007-05-31 | 2007-05-31 | Method and apparatus for MPI program optimization |
PCT/CN2007/001747 WO2008144960A1 (en) | 2007-05-31 | 2007-05-31 | Method and apparatus for mpi program optimization |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/CN2007/001747 WO2008144960A1 (en) | 2007-05-31 | 2007-05-31 | Method and apparatus for mpi program optimization |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2008144960A1 true WO2008144960A1 (en) | 2008-12-04 |
Family
ID=40074543
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2007/001747 WO2008144960A1 (en) | 2007-05-31 | 2007-05-31 | Method and apparatus for mpi program optimization |
Country Status (2)
Country | Link |
---|---|
US (1) | US8312227B2 (en) |
WO (1) | WO2008144960A1 (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8312227B2 (en) | 2007-05-31 | 2012-11-13 | Intel Corporation | Method and apparatus for MPI program optimization |
CN111625954A (en) * | 2020-05-22 | 2020-09-04 | 中国地质大学(武汉) | Parallel optimization method and system for rainfall type landslide model TRIGRS |
Families Citing this family (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8234652B2 (en) | 2007-08-28 | 2012-07-31 | International Business Machines Corporation | Performing setup operations for receiving different amounts of data while processors are performing message passing interface tasks |
US8312464B2 (en) * | 2007-08-28 | 2012-11-13 | International Business Machines Corporation | Hardware based dynamic load balancing of message passing interface tasks by modifying tasks |
US8127300B2 (en) * | 2007-08-28 | 2012-02-28 | International Business Machines Corporation | Hardware based dynamic load balancing of message passing interface tasks |
US20090064166A1 (en) * | 2007-08-28 | 2009-03-05 | Arimilli Lakshminarayana B | System and Method for Hardware Based Dynamic Load Balancing of Message Passing Interface Tasks |
US8108876B2 (en) | 2007-08-28 | 2012-01-31 | International Business Machines Corporation | Modifying an operation of one or more processors executing message passing interface tasks |
US8336050B2 (en) * | 2009-08-31 | 2012-12-18 | Red Hat, Inc. | Shared memory inter-process communication of virtual machines using virtual synchrony |
CN102426536B (en) * | 2011-10-26 | 2014-04-09 | 深圳市亚特尔科技有限公司 | Multitask data communication implementation method and system |
US10936432B1 (en) | 2014-09-24 | 2021-03-02 | Amazon Technologies, Inc. | Fault-tolerant parallel computation |
CN107180153A (en) * | 2016-03-11 | 2017-09-19 | 中国石油化工股份有限公司 | The method and system of full waveform inversion is realized using MPI |
US10963323B2 (en) * | 2018-10-25 | 2021-03-30 | Sangyung University Industry-Academy Cooperation Foundation | Method and apparatus for transformation of MPI programs for memory centric computers |
CN111061652B (en) * | 2019-12-18 | 2021-12-31 | 中山大学 | Nonvolatile memory management method and system based on MPI-IO middleware |
US11836549B2 (en) | 2020-10-15 | 2023-12-05 | Advanced Micro Devices, Inc. | Fast block-based parallel message passing interface transpose |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5555396A (en) * | 1994-12-22 | 1996-09-10 | Unisys Corporation | Hierarchical queuing in a system architecture for improved message passing and process synchronization |
US20040107419A1 (en) * | 2002-12-03 | 2004-06-03 | International Business Machines Corporation | Efficient shared memory transport in a distributed data processing environment |
US6799317B1 (en) * | 2000-06-27 | 2004-09-28 | International Business Machines Corporation | Interrupt mechanism for shared memory message passing |
WO2005033882A2 (en) * | 2003-09-29 | 2005-04-14 | Verari Systems Software, Inc. | System and method for high performance message passing |
US20060080668A1 (en) * | 2004-10-12 | 2006-04-13 | International Business Machines Corporation | Facilitating intra-node data transfer in collective communications |
Family Cites Families (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20030066056A1 (en) * | 2001-09-28 | 2003-04-03 | Petersen Paul M. | Method and apparatus for accessing thread-privatized global storage objects |
US7293267B1 (en) * | 2003-12-22 | 2007-11-06 | Sun Microsystems Inc | System and method for performing speculative initialization of application models for a cloned runtime system process |
US7454477B2 (en) * | 2005-05-16 | 2008-11-18 | Microsoft Corporation | Zero-copy transfer of memory between address spaces |
US20070192267A1 (en) * | 2006-02-10 | 2007-08-16 | Numenta, Inc. | Architecture of a hierarchical temporal memory based system |
US8056063B2 (en) * | 2006-08-07 | 2011-11-08 | International Business Machines Corporation | Method and apparatus minimizing code duplication in a statically typeable language program |
GB2443277B (en) * | 2006-10-24 | 2011-05-18 | Advanced Risc Mach Ltd | Performing diagnostics operations upon an asymmetric multiprocessor apparatus |
US20080148095A1 (en) * | 2006-12-14 | 2008-06-19 | Motorola, Inc. | Automated memory recovery in a zero copy messaging system |
US8341609B2 (en) * | 2007-01-26 | 2012-12-25 | Oracle International Corporation | Code generation in the presence of paged memory |
WO2008144960A1 (en) | 2007-05-31 | 2008-12-04 | Intel Coporation | Method and apparatus for mpi program optimization |
US8108876B2 (en) * | 2007-08-28 | 2012-01-31 | International Business Machines Corporation | Modifying an operation of one or more processors executing message passing interface tasks |
US8347038B2 (en) * | 2008-03-31 | 2013-01-01 | Intel Corporation | Optimizing memory copy routine selection for message passing in a multicore architecture |
US20090307669A1 (en) * | 2008-06-06 | 2009-12-10 | Garst Jr Gerald Blaine | Memory management for closures |
-
2007
- 2007-05-31 WO PCT/CN2007/001747 patent/WO2008144960A1/en active Application Filing
- 2007-05-31 US US12/598,597 patent/US8312227B2/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5555396A (en) * | 1994-12-22 | 1996-09-10 | Unisys Corporation | Hierarchical queuing in a system architecture for improved message passing and process synchronization |
US6799317B1 (en) * | 2000-06-27 | 2004-09-28 | International Business Machines Corporation | Interrupt mechanism for shared memory message passing |
US20040107419A1 (en) * | 2002-12-03 | 2004-06-03 | International Business Machines Corporation | Efficient shared memory transport in a distributed data processing environment |
WO2005033882A2 (en) * | 2003-09-29 | 2005-04-14 | Verari Systems Software, Inc. | System and method for high performance message passing |
US20060080668A1 (en) * | 2004-10-12 | 2006-04-13 | International Business Machines Corporation | Facilitating intra-node data transfer in collective communications |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8312227B2 (en) | 2007-05-31 | 2012-11-13 | Intel Corporation | Method and apparatus for MPI program optimization |
CN111625954A (en) * | 2020-05-22 | 2020-09-04 | 中国地质大学(武汉) | Parallel optimization method and system for rainfall type landslide model TRIGRS |
CN111625954B (en) * | 2020-05-22 | 2023-10-27 | 中国地质大学(武汉) | Parallel optimization method and system for rainfall landslide model TRIGRS |
Also Published As
Publication number | Publication date |
---|---|
US20100161911A1 (en) | 2010-06-24 |
US8312227B2 (en) | 2012-11-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US8312227B2 (en) | Method and apparatus for MPI program optimization | |
JP6342970B2 (en) | Read and write monitoring attributes in transactional memory (TM) systems | |
US10169013B2 (en) | Arranging binary code based on call graph partitioning | |
KR101291016B1 (en) | Registering a user-handler in hardware for transactional memory event handling | |
US9727471B2 (en) | Method and apparatus for stream buffer management instructions | |
RU2501071C2 (en) | Late lock acquire mechanism for hardware lock elision (hle) | |
US9513886B2 (en) | Heap data management for limited local memory(LLM) multi-core processors | |
CN108733415B (en) | Method and device for supporting vector random access | |
CN102834813B (en) | For the renewal processor of multi-channel high-speed buffer memory | |
CN108268385B (en) | Optimized caching agent with integrated directory cache | |
US20080141268A1 (en) | Utility function execution using scout threads | |
US20110320786A1 (en) | Dynamically Rewriting Branch Instructions in Response to Cache Line Eviction | |
US10031697B2 (en) | Random-access disjoint concurrent sparse writes to heterogeneous buffers | |
CN102667714B (en) | Support the method and system that the function provided by the resource outside operating system environment is provided | |
US8726248B2 (en) | Method and apparatus for enregistering memory locations | |
US20150234687A1 (en) | Thread migration across cores of a multi-core processor | |
US8490071B2 (en) | Shared prefetching to reduce execution skew in multi-threaded systems | |
US20190187964A1 (en) | Method and Apparatus for Compiler Driven Bank Conflict Avoidance | |
US8949777B2 (en) | Methods and systems for mapping a function pointer to the device code | |
US7562204B1 (en) | Identifying and relocating relocatable kernel memory allocations in kernel non-relocatable memory | |
US9244828B2 (en) | Allocating memory and using the allocated memory in a workgroup in a dispatched data parallel kernel | |
US20220318015A1 (en) | Enforcing data placement requirements via address bit swapping | |
US12086622B2 (en) | Optimizing virtual machine scheduling on non-uniform cache access (NUCA) systems | |
US20050251795A1 (en) | Method, system, and program for optimizing code | |
CN118069373B (en) | Electronic component, electronic device and memory management method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 07721320 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
WWE | Wipo information: entry into national phase |
Ref document number: 12598597 Country of ref document: US |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 07721320 Country of ref document: EP Kind code of ref document: A1 |